g_mirror.c revision 163836
1/*- 2 * Copyright (c) 2004-2006 Pawel Jakub Dawidek <pjd@FreeBSD.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27#include <sys/cdefs.h> 28__FBSDID("$FreeBSD: head/sys/geom/mirror/g_mirror.c 163836 2006-10-31 21:23:51Z pjd $"); 29 30#include <sys/param.h> 31#include <sys/systm.h> 32#include <sys/kernel.h> 33#include <sys/module.h> 34#include <sys/limits.h> 35#include <sys/lock.h> 36#include <sys/mutex.h> 37#include <sys/bio.h> 38#include <sys/sysctl.h> 39#include <sys/malloc.h> 40#include <sys/eventhandler.h> 41#include <vm/uma.h> 42#include <geom/geom.h> 43#include <sys/proc.h> 44#include <sys/kthread.h> 45#include <sys/sched.h> 46#include <geom/mirror/g_mirror.h> 47 48 49static MALLOC_DEFINE(M_MIRROR, "mirror_data", "GEOM_MIRROR Data"); 50 51SYSCTL_DECL(_kern_geom); 52SYSCTL_NODE(_kern_geom, OID_AUTO, mirror, CTLFLAG_RW, 0, "GEOM_MIRROR stuff"); 53u_int g_mirror_debug = 0; 54TUNABLE_INT("kern.geom.mirror.debug", &g_mirror_debug); 55SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RW, &g_mirror_debug, 0, 56 "Debug level"); 57static u_int g_mirror_timeout = 4; 58TUNABLE_INT("kern.geom.mirror.timeout", &g_mirror_timeout); 59SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RW, &g_mirror_timeout, 60 0, "Time to wait on all mirror components"); 61static u_int g_mirror_idletime = 5; 62TUNABLE_INT("kern.geom.mirror.idletime", &g_mirror_idletime); 63SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, idletime, CTLFLAG_RW, 64 &g_mirror_idletime, 0, "Mark components as clean when idling"); 65static u_int g_mirror_disconnect_on_failure = 1; 66TUNABLE_INT("kern.geom.mirror.disconnect_on_failure", 67 &g_mirror_disconnect_on_failure); 68SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, disconnect_on_failure, CTLFLAG_RW, 69 &g_mirror_disconnect_on_failure, 0, "Disconnect component on I/O failure."); 70static u_int g_mirror_syncreqs = 2; 71TUNABLE_INT("kern.geom.mirror.sync_requests", &g_mirror_syncreqs); 72SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, sync_requests, CTLFLAG_RDTUN, 73 &g_mirror_syncreqs, 0, "Parallel synchronization I/O requests."); 74 75#define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 76 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 77 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 78 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 79} while (0) 80 81static eventhandler_tag g_mirror_pre_sync = NULL; 82 83static int g_mirror_destroy_geom(struct gctl_req *req, struct g_class *mp, 84 struct g_geom *gp); 85static g_taste_t g_mirror_taste; 86static void g_mirror_init(struct g_class *mp); 87static void g_mirror_fini(struct g_class *mp); 88 89struct g_class g_mirror_class = { 90 .name = G_MIRROR_CLASS_NAME, 91 .version = G_VERSION, 92 .ctlreq = g_mirror_config, 93 .taste = g_mirror_taste, 94 .destroy_geom = g_mirror_destroy_geom, 95 .init = g_mirror_init, 96 .fini = g_mirror_fini 97}; 98 99 100static void g_mirror_destroy_provider(struct g_mirror_softc *sc); 101static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 102static void g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force); 103static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 104 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 105static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 106static void g_mirror_register_request(struct bio *bp); 107static void g_mirror_sync_release(struct g_mirror_softc *sc); 108 109 110static const char * 111g_mirror_disk_state2str(int state) 112{ 113 114 switch (state) { 115 case G_MIRROR_DISK_STATE_NONE: 116 return ("NONE"); 117 case G_MIRROR_DISK_STATE_NEW: 118 return ("NEW"); 119 case G_MIRROR_DISK_STATE_ACTIVE: 120 return ("ACTIVE"); 121 case G_MIRROR_DISK_STATE_STALE: 122 return ("STALE"); 123 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 124 return ("SYNCHRONIZING"); 125 case G_MIRROR_DISK_STATE_DISCONNECTED: 126 return ("DISCONNECTED"); 127 case G_MIRROR_DISK_STATE_DESTROY: 128 return ("DESTROY"); 129 default: 130 return ("INVALID"); 131 } 132} 133 134static const char * 135g_mirror_device_state2str(int state) 136{ 137 138 switch (state) { 139 case G_MIRROR_DEVICE_STATE_STARTING: 140 return ("STARTING"); 141 case G_MIRROR_DEVICE_STATE_RUNNING: 142 return ("RUNNING"); 143 default: 144 return ("INVALID"); 145 } 146} 147 148static const char * 149g_mirror_get_diskname(struct g_mirror_disk *disk) 150{ 151 152 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 153 return ("[unknown]"); 154 return (disk->d_name); 155} 156 157/* 158 * --- Events handling functions --- 159 * Events in geom_mirror are used to maintain disks and device status 160 * from one thread to simplify locking. 161 */ 162static void 163g_mirror_event_free(struct g_mirror_event *ep) 164{ 165 166 free(ep, M_MIRROR); 167} 168 169int 170g_mirror_event_send(void *arg, int state, int flags) 171{ 172 struct g_mirror_softc *sc; 173 struct g_mirror_disk *disk; 174 struct g_mirror_event *ep; 175 int error; 176 177 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 178 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 179 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 180 disk = NULL; 181 sc = arg; 182 } else { 183 disk = arg; 184 sc = disk->d_softc; 185 } 186 ep->e_disk = disk; 187 ep->e_state = state; 188 ep->e_flags = flags; 189 ep->e_error = 0; 190 mtx_lock(&sc->sc_events_mtx); 191 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 192 mtx_unlock(&sc->sc_events_mtx); 193 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 194 mtx_lock(&sc->sc_queue_mtx); 195 wakeup(sc); 196 mtx_unlock(&sc->sc_queue_mtx); 197 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 198 return (0); 199 sx_assert(&sc->sc_lock, SX_XLOCKED); 200 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 201 sx_xunlock(&sc->sc_lock); 202 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 203 mtx_lock(&sc->sc_events_mtx); 204 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 205 hz * 5); 206 } 207 error = ep->e_error; 208 g_mirror_event_free(ep); 209 sx_xlock(&sc->sc_lock); 210 return (error); 211} 212 213static struct g_mirror_event * 214g_mirror_event_get(struct g_mirror_softc *sc) 215{ 216 struct g_mirror_event *ep; 217 218 mtx_lock(&sc->sc_events_mtx); 219 ep = TAILQ_FIRST(&sc->sc_events); 220 mtx_unlock(&sc->sc_events_mtx); 221 return (ep); 222} 223 224static void 225g_mirror_event_remove(struct g_mirror_softc *sc, struct g_mirror_event *ep) 226{ 227 228 mtx_lock(&sc->sc_events_mtx); 229 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 230 mtx_unlock(&sc->sc_events_mtx); 231} 232 233static void 234g_mirror_event_cancel(struct g_mirror_disk *disk) 235{ 236 struct g_mirror_softc *sc; 237 struct g_mirror_event *ep, *tmpep; 238 239 sc = disk->d_softc; 240 sx_assert(&sc->sc_lock, SX_XLOCKED); 241 242 mtx_lock(&sc->sc_events_mtx); 243 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 244 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 245 continue; 246 if (ep->e_disk != disk) 247 continue; 248 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 249 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 250 g_mirror_event_free(ep); 251 else { 252 ep->e_error = ECANCELED; 253 wakeup(ep); 254 } 255 } 256 mtx_unlock(&sc->sc_events_mtx); 257} 258 259/* 260 * Return the number of disks in given state. 261 * If state is equal to -1, count all connected disks. 262 */ 263u_int 264g_mirror_ndisks(struct g_mirror_softc *sc, int state) 265{ 266 struct g_mirror_disk *disk; 267 u_int n = 0; 268 269 sx_assert(&sc->sc_lock, SX_LOCKED); 270 271 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 272 if (state == -1 || disk->d_state == state) 273 n++; 274 } 275 return (n); 276} 277 278/* 279 * Find a disk in mirror by its disk ID. 280 */ 281static struct g_mirror_disk * 282g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 283{ 284 struct g_mirror_disk *disk; 285 286 sx_assert(&sc->sc_lock, SX_XLOCKED); 287 288 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 289 if (disk->d_id == id) 290 return (disk); 291 } 292 return (NULL); 293} 294 295static u_int 296g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 297{ 298 struct bio *bp; 299 u_int nreqs = 0; 300 301 mtx_lock(&sc->sc_queue_mtx); 302 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 303 if (bp->bio_from == cp) 304 nreqs++; 305 } 306 mtx_unlock(&sc->sc_queue_mtx); 307 return (nreqs); 308} 309 310static int 311g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 312{ 313 314 if (cp->index > 0) { 315 G_MIRROR_DEBUG(2, 316 "I/O requests for %s exist, can't destroy it now.", 317 cp->provider->name); 318 return (1); 319 } 320 if (g_mirror_nrequests(sc, cp) > 0) { 321 G_MIRROR_DEBUG(2, 322 "I/O requests for %s in queue, can't destroy it now.", 323 cp->provider->name); 324 return (1); 325 } 326 return (0); 327} 328 329static void 330g_mirror_destroy_consumer(void *arg, int flags __unused) 331{ 332 struct g_consumer *cp; 333 334 g_topology_assert(); 335 336 cp = arg; 337 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 338 g_detach(cp); 339 g_destroy_consumer(cp); 340} 341 342static void 343g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 344{ 345 struct g_provider *pp; 346 int retaste_wait; 347 348 g_topology_assert(); 349 350 cp->private = NULL; 351 if (g_mirror_is_busy(sc, cp)) 352 return; 353 pp = cp->provider; 354 retaste_wait = 0; 355 if (cp->acw == 1) { 356 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 357 retaste_wait = 1; 358 } 359 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 360 -cp->acw, -cp->ace, 0); 361 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 362 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 363 if (retaste_wait) { 364 /* 365 * After retaste event was send (inside g_access()), we can send 366 * event to detach and destroy consumer. 367 * A class, which has consumer to the given provider connected 368 * will not receive retaste event for the provider. 369 * This is the way how I ignore retaste events when I close 370 * consumers opened for write: I detach and destroy consumer 371 * after retaste event is sent. 372 */ 373 g_post_event(g_mirror_destroy_consumer, cp, M_WAITOK, NULL); 374 return; 375 } 376 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", pp->name); 377 g_detach(cp); 378 g_destroy_consumer(cp); 379} 380 381static int 382g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 383{ 384 struct g_consumer *cp; 385 int error; 386 387 g_topology_assert_not(); 388 KASSERT(disk->d_consumer == NULL, 389 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 390 391 g_topology_lock(); 392 cp = g_new_consumer(disk->d_softc->sc_geom); 393 error = g_attach(cp, pp); 394 if (error != 0) { 395 g_destroy_consumer(cp); 396 g_topology_unlock(); 397 return (error); 398 } 399 error = g_access(cp, 1, 1, 1); 400 if (error != 0) { 401 g_detach(cp); 402 g_destroy_consumer(cp); 403 g_topology_unlock(); 404 G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).", 405 pp->name, error); 406 return (error); 407 } 408 g_topology_unlock(); 409 disk->d_consumer = cp; 410 disk->d_consumer->private = disk; 411 disk->d_consumer->index = 0; 412 413 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 414 return (0); 415} 416 417static void 418g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 419{ 420 421 g_topology_assert(); 422 423 if (cp == NULL) 424 return; 425 if (cp->provider != NULL) 426 g_mirror_kill_consumer(sc, cp); 427 else 428 g_destroy_consumer(cp); 429} 430 431/* 432 * Initialize disk. This means allocate memory, create consumer, attach it 433 * to the provider and open access (r1w1e1) to it. 434 */ 435static struct g_mirror_disk * 436g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 437 struct g_mirror_metadata *md, int *errorp) 438{ 439 struct g_mirror_disk *disk; 440 int error; 441 442 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 443 if (disk == NULL) { 444 error = ENOMEM; 445 goto fail; 446 } 447 disk->d_softc = sc; 448 error = g_mirror_connect_disk(disk, pp); 449 if (error != 0) 450 goto fail; 451 disk->d_id = md->md_did; 452 disk->d_state = G_MIRROR_DISK_STATE_NONE; 453 disk->d_priority = md->md_priority; 454 disk->d_delay.sec = 0; 455 disk->d_delay.frac = 0; 456 binuptime(&disk->d_last_used); 457 disk->d_flags = md->md_dflags; 458 if (md->md_provider[0] != '\0') 459 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 460 disk->d_sync.ds_consumer = NULL; 461 disk->d_sync.ds_offset = md->md_sync_offset; 462 disk->d_sync.ds_offset_done = md->md_sync_offset; 463 disk->d_genid = md->md_genid; 464 disk->d_sync.ds_syncid = md->md_syncid; 465 if (errorp != NULL) 466 *errorp = 0; 467 return (disk); 468fail: 469 if (errorp != NULL) 470 *errorp = error; 471 if (disk != NULL) 472 free(disk, M_MIRROR); 473 return (NULL); 474} 475 476static void 477g_mirror_destroy_disk(struct g_mirror_disk *disk) 478{ 479 struct g_mirror_softc *sc; 480 481 g_topology_assert_not(); 482 sc = disk->d_softc; 483 sx_assert(&sc->sc_lock, SX_XLOCKED); 484 485 LIST_REMOVE(disk, d_next); 486 g_mirror_event_cancel(disk); 487 if (sc->sc_hint == disk) 488 sc->sc_hint = NULL; 489 switch (disk->d_state) { 490 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 491 g_mirror_sync_stop(disk, 1); 492 /* FALLTHROUGH */ 493 case G_MIRROR_DISK_STATE_NEW: 494 case G_MIRROR_DISK_STATE_STALE: 495 case G_MIRROR_DISK_STATE_ACTIVE: 496 g_topology_lock(); 497 g_mirror_disconnect_consumer(sc, disk->d_consumer); 498 g_topology_unlock(); 499 free(disk, M_MIRROR); 500 break; 501 default: 502 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 503 g_mirror_get_diskname(disk), 504 g_mirror_disk_state2str(disk->d_state))); 505 } 506} 507 508static void 509g_mirror_destroy_device(struct g_mirror_softc *sc) 510{ 511 struct g_mirror_disk *disk; 512 struct g_mirror_event *ep; 513 struct g_geom *gp; 514 struct g_consumer *cp, *tmpcp; 515 516 g_topology_assert_not(); 517 sx_assert(&sc->sc_lock, SX_XLOCKED); 518 519 gp = sc->sc_geom; 520 if (sc->sc_provider != NULL) 521 g_mirror_destroy_provider(sc); 522 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 523 disk = LIST_FIRST(&sc->sc_disks)) { 524 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 525 g_mirror_update_metadata(disk); 526 g_mirror_destroy_disk(disk); 527 } 528 while ((ep = g_mirror_event_get(sc)) != NULL) { 529 g_mirror_event_remove(sc, ep); 530 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 531 g_mirror_event_free(ep); 532 else { 533 ep->e_error = ECANCELED; 534 ep->e_flags |= G_MIRROR_EVENT_DONE; 535 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 536 mtx_lock(&sc->sc_events_mtx); 537 wakeup(ep); 538 mtx_unlock(&sc->sc_events_mtx); 539 } 540 } 541 callout_drain(&sc->sc_callout); 542 543 g_topology_lock(); 544 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 545 g_mirror_disconnect_consumer(sc, cp); 546 } 547 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 548 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 549 g_wither_geom(gp, ENXIO); 550 g_topology_unlock(); 551 mtx_destroy(&sc->sc_queue_mtx); 552 mtx_destroy(&sc->sc_events_mtx); 553 sx_xunlock(&sc->sc_lock); 554 sx_destroy(&sc->sc_lock); 555} 556 557static void 558g_mirror_orphan(struct g_consumer *cp) 559{ 560 struct g_mirror_disk *disk; 561 562 g_topology_assert(); 563 564 disk = cp->private; 565 if (disk == NULL) 566 return; 567 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 568 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 569 G_MIRROR_EVENT_DONTWAIT); 570} 571 572/* 573 * Function should return the next active disk on the list. 574 * It is possible that it will be the same disk as given. 575 * If there are no active disks on list, NULL is returned. 576 */ 577static __inline struct g_mirror_disk * 578g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 579{ 580 struct g_mirror_disk *dp; 581 582 for (dp = LIST_NEXT(disk, d_next); dp != disk; 583 dp = LIST_NEXT(dp, d_next)) { 584 if (dp == NULL) 585 dp = LIST_FIRST(&sc->sc_disks); 586 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 587 break; 588 } 589 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 590 return (NULL); 591 return (dp); 592} 593 594static struct g_mirror_disk * 595g_mirror_get_disk(struct g_mirror_softc *sc) 596{ 597 struct g_mirror_disk *disk; 598 599 if (sc->sc_hint == NULL) { 600 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 601 if (sc->sc_hint == NULL) 602 return (NULL); 603 } 604 disk = sc->sc_hint; 605 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 606 disk = g_mirror_find_next(sc, disk); 607 if (disk == NULL) 608 return (NULL); 609 } 610 sc->sc_hint = g_mirror_find_next(sc, disk); 611 return (disk); 612} 613 614static int 615g_mirror_write_metadata(struct g_mirror_disk *disk, 616 struct g_mirror_metadata *md) 617{ 618 struct g_mirror_softc *sc; 619 struct g_consumer *cp; 620 off_t offset, length; 621 u_char *sector; 622 int error = 0; 623 624 g_topology_assert_not(); 625 sc = disk->d_softc; 626 sx_assert(&sc->sc_lock, SX_LOCKED); 627 628 cp = disk->d_consumer; 629 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 630 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 631 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 632 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 633 cp->acw, cp->ace)); 634 length = cp->provider->sectorsize; 635 offset = cp->provider->mediasize - length; 636 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 637 if (md != NULL) 638 mirror_metadata_encode(md, sector); 639 error = g_write_data(cp, offset, sector, length); 640 free(sector, M_MIRROR); 641 if (error != 0) { 642 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 643 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 644 G_MIRROR_DEBUG(0, "Cannot write metadata on %s " 645 "(device=%s, error=%d).", 646 g_mirror_get_diskname(disk), sc->sc_name, error); 647 } else { 648 G_MIRROR_DEBUG(1, "Cannot write metadata on %s " 649 "(device=%s, error=%d).", 650 g_mirror_get_diskname(disk), sc->sc_name, error); 651 } 652 if (g_mirror_disconnect_on_failure && 653 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) { 654 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 655 g_mirror_event_send(disk, 656 G_MIRROR_DISK_STATE_DISCONNECTED, 657 G_MIRROR_EVENT_DONTWAIT); 658 } 659 } 660 return (error); 661} 662 663static int 664g_mirror_clear_metadata(struct g_mirror_disk *disk) 665{ 666 int error; 667 668 g_topology_assert_not(); 669 sx_assert(&disk->d_softc->sc_lock, SX_LOCKED); 670 671 error = g_mirror_write_metadata(disk, NULL); 672 if (error == 0) { 673 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 674 g_mirror_get_diskname(disk)); 675 } else { 676 G_MIRROR_DEBUG(0, 677 "Cannot clear metadata on disk %s (error=%d).", 678 g_mirror_get_diskname(disk), error); 679 } 680 return (error); 681} 682 683void 684g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 685 struct g_mirror_metadata *md) 686{ 687 688 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 689 md->md_version = G_MIRROR_VERSION; 690 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 691 md->md_mid = sc->sc_id; 692 md->md_all = sc->sc_ndisks; 693 md->md_slice = sc->sc_slice; 694 md->md_balance = sc->sc_balance; 695 md->md_genid = sc->sc_genid; 696 md->md_mediasize = sc->sc_mediasize; 697 md->md_sectorsize = sc->sc_sectorsize; 698 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 699 bzero(md->md_provider, sizeof(md->md_provider)); 700 if (disk == NULL) { 701 md->md_did = arc4random(); 702 md->md_priority = 0; 703 md->md_syncid = 0; 704 md->md_dflags = 0; 705 md->md_sync_offset = 0; 706 md->md_provsize = 0; 707 } else { 708 md->md_did = disk->d_id; 709 md->md_priority = disk->d_priority; 710 md->md_syncid = disk->d_sync.ds_syncid; 711 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 712 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 713 md->md_sync_offset = disk->d_sync.ds_offset_done; 714 else 715 md->md_sync_offset = 0; 716 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 717 strlcpy(md->md_provider, 718 disk->d_consumer->provider->name, 719 sizeof(md->md_provider)); 720 } 721 md->md_provsize = disk->d_consumer->provider->mediasize; 722 } 723} 724 725void 726g_mirror_update_metadata(struct g_mirror_disk *disk) 727{ 728 struct g_mirror_softc *sc; 729 struct g_mirror_metadata md; 730 int error; 731 732 g_topology_assert_not(); 733 sc = disk->d_softc; 734 sx_assert(&sc->sc_lock, SX_LOCKED); 735 736 g_mirror_fill_metadata(sc, disk, &md); 737 error = g_mirror_write_metadata(disk, &md); 738 if (error == 0) { 739 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 740 g_mirror_get_diskname(disk)); 741 } else { 742 G_MIRROR_DEBUG(0, 743 "Cannot update metadata on disk %s (error=%d).", 744 g_mirror_get_diskname(disk), error); 745 } 746} 747 748static void 749g_mirror_bump_syncid(struct g_mirror_softc *sc) 750{ 751 struct g_mirror_disk *disk; 752 753 g_topology_assert_not(); 754 sx_assert(&sc->sc_lock, SX_XLOCKED); 755 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 756 ("%s called with no active disks (device=%s).", __func__, 757 sc->sc_name)); 758 759 sc->sc_syncid++; 760 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 761 sc->sc_syncid); 762 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 763 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 764 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 765 disk->d_sync.ds_syncid = sc->sc_syncid; 766 g_mirror_update_metadata(disk); 767 } 768 } 769} 770 771static void 772g_mirror_bump_genid(struct g_mirror_softc *sc) 773{ 774 struct g_mirror_disk *disk; 775 776 g_topology_assert_not(); 777 sx_assert(&sc->sc_lock, SX_XLOCKED); 778 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 779 ("%s called with no active disks (device=%s).", __func__, 780 sc->sc_name)); 781 782 sc->sc_genid++; 783 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 784 sc->sc_genid); 785 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 786 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 787 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 788 disk->d_genid = sc->sc_genid; 789 g_mirror_update_metadata(disk); 790 } 791 } 792} 793 794static int 795g_mirror_idle(struct g_mirror_softc *sc, int acw) 796{ 797 struct g_mirror_disk *disk; 798 int timeout; 799 800 g_topology_assert_not(); 801 sx_assert(&sc->sc_lock, SX_XLOCKED); 802 803 if (sc->sc_provider == NULL) 804 return (0); 805 if (sc->sc_idle) 806 return (0); 807 if (sc->sc_writes > 0) 808 return (0); 809 if (acw > 0 || (acw == -1 && sc->sc_provider->acw > 0)) { 810 timeout = g_mirror_idletime - (time_uptime - sc->sc_last_write); 811 if (timeout > 0) 812 return (timeout); 813 } 814 sc->sc_idle = 1; 815 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 816 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 817 continue; 818 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 819 g_mirror_get_diskname(disk), sc->sc_name); 820 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 821 g_mirror_update_metadata(disk); 822 } 823 return (0); 824} 825 826static void 827g_mirror_unidle(struct g_mirror_softc *sc) 828{ 829 struct g_mirror_disk *disk; 830 831 g_topology_assert_not(); 832 sx_assert(&sc->sc_lock, SX_XLOCKED); 833 834 sc->sc_idle = 0; 835 sc->sc_last_write = time_uptime; 836 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 837 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 838 continue; 839 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 840 g_mirror_get_diskname(disk), sc->sc_name); 841 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 842 g_mirror_update_metadata(disk); 843 } 844} 845 846static __inline int 847bintime_cmp(struct bintime *bt1, struct bintime *bt2) 848{ 849 850 if (bt1->sec < bt2->sec) 851 return (-1); 852 else if (bt1->sec > bt2->sec) 853 return (1); 854 if (bt1->frac < bt2->frac) 855 return (-1); 856 else if (bt1->frac > bt2->frac) 857 return (1); 858 return (0); 859} 860 861static void 862g_mirror_update_delay(struct g_mirror_disk *disk, struct bio *bp) 863{ 864 865 if (disk->d_softc->sc_balance != G_MIRROR_BALANCE_LOAD) 866 return; 867 binuptime(&disk->d_delay); 868 bintime_sub(&disk->d_delay, &bp->bio_t0); 869} 870 871static void 872g_mirror_done(struct bio *bp) 873{ 874 struct g_mirror_softc *sc; 875 876 sc = bp->bio_from->geom->softc; 877 bp->bio_cflags = G_MIRROR_BIO_FLAG_REGULAR; 878 mtx_lock(&sc->sc_queue_mtx); 879 bioq_disksort(&sc->sc_queue, bp); 880 wakeup(sc); 881 mtx_unlock(&sc->sc_queue_mtx); 882} 883 884static void 885g_mirror_regular_request(struct bio *bp) 886{ 887 struct g_mirror_softc *sc; 888 struct g_mirror_disk *disk; 889 struct bio *pbp; 890 891 g_topology_assert_not(); 892 893 pbp = bp->bio_parent; 894 sc = pbp->bio_to->geom->softc; 895 bp->bio_from->index--; 896 if (bp->bio_cmd == BIO_WRITE) 897 sc->sc_writes--; 898 disk = bp->bio_from->private; 899 if (disk == NULL) { 900 g_topology_lock(); 901 g_mirror_kill_consumer(sc, bp->bio_from); 902 g_topology_unlock(); 903 } else { 904 g_mirror_update_delay(disk, bp); 905 } 906 907 pbp->bio_inbed++; 908 KASSERT(pbp->bio_inbed <= pbp->bio_children, 909 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 910 pbp->bio_children)); 911 if (bp->bio_error == 0 && pbp->bio_error == 0) { 912 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 913 g_destroy_bio(bp); 914 if (pbp->bio_children == pbp->bio_inbed) { 915 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 916 pbp->bio_completed = pbp->bio_length; 917 if (pbp->bio_cmd == BIO_WRITE) { 918 bioq_remove(&sc->sc_inflight, pbp); 919 /* Release delayed sync requests if possible. */ 920 g_mirror_sync_release(sc); 921 } 922 g_io_deliver(pbp, pbp->bio_error); 923 } 924 return; 925 } else if (bp->bio_error != 0) { 926 if (pbp->bio_error == 0) 927 pbp->bio_error = bp->bio_error; 928 if (disk != NULL) { 929 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 930 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 931 G_MIRROR_LOGREQ(0, bp, 932 "Request failed (error=%d).", 933 bp->bio_error); 934 } else { 935 G_MIRROR_LOGREQ(1, bp, 936 "Request failed (error=%d).", 937 bp->bio_error); 938 } 939 if (g_mirror_disconnect_on_failure && 940 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) 941 { 942 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 943 g_mirror_event_send(disk, 944 G_MIRROR_DISK_STATE_DISCONNECTED, 945 G_MIRROR_EVENT_DONTWAIT); 946 } 947 } 948 switch (pbp->bio_cmd) { 949 case BIO_DELETE: 950 case BIO_WRITE: 951 pbp->bio_inbed--; 952 pbp->bio_children--; 953 break; 954 } 955 } 956 g_destroy_bio(bp); 957 958 switch (pbp->bio_cmd) { 959 case BIO_READ: 960 if (pbp->bio_inbed < pbp->bio_children) 961 break; 962 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 1) 963 g_io_deliver(pbp, pbp->bio_error); 964 else { 965 pbp->bio_error = 0; 966 mtx_lock(&sc->sc_queue_mtx); 967 bioq_disksort(&sc->sc_queue, pbp); 968 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 969 wakeup(sc); 970 mtx_unlock(&sc->sc_queue_mtx); 971 } 972 break; 973 case BIO_DELETE: 974 case BIO_WRITE: 975 if (pbp->bio_children == 0) { 976 /* 977 * All requests failed. 978 */ 979 } else if (pbp->bio_inbed < pbp->bio_children) { 980 /* Do nothing. */ 981 break; 982 } else if (pbp->bio_children == pbp->bio_inbed) { 983 /* Some requests succeeded. */ 984 pbp->bio_error = 0; 985 pbp->bio_completed = pbp->bio_length; 986 } 987 bioq_remove(&sc->sc_inflight, pbp); 988 /* Release delayed sync requests if possible. */ 989 g_mirror_sync_release(sc); 990 g_io_deliver(pbp, pbp->bio_error); 991 break; 992 default: 993 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 994 break; 995 } 996} 997 998static void 999g_mirror_sync_done(struct bio *bp) 1000{ 1001 struct g_mirror_softc *sc; 1002 1003 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 1004 sc = bp->bio_from->geom->softc; 1005 bp->bio_cflags = G_MIRROR_BIO_FLAG_SYNC; 1006 mtx_lock(&sc->sc_queue_mtx); 1007 bioq_disksort(&sc->sc_queue, bp); 1008 wakeup(sc); 1009 mtx_unlock(&sc->sc_queue_mtx); 1010} 1011 1012static void 1013g_mirror_kernel_dump(struct bio *bp) 1014{ 1015 struct g_mirror_softc *sc; 1016 struct g_mirror_disk *disk; 1017 struct bio *cbp; 1018 struct g_kerneldump *gkd; 1019 1020 /* 1021 * We configure dumping to the first component, because this component 1022 * will be used for reading with 'prefer' balance algorithm. 1023 * If the component with the higest priority is currently disconnected 1024 * we will not be able to read the dump after the reboot if it will be 1025 * connected and synchronized later. Can we do something better? 1026 */ 1027 sc = bp->bio_to->geom->softc; 1028 disk = LIST_FIRST(&sc->sc_disks); 1029 1030 gkd = (struct g_kerneldump *)bp->bio_data; 1031 if (gkd->length > bp->bio_to->mediasize) 1032 gkd->length = bp->bio_to->mediasize; 1033 cbp = g_clone_bio(bp); 1034 if (cbp == NULL) { 1035 g_io_deliver(bp, ENOMEM); 1036 return; 1037 } 1038 cbp->bio_done = g_std_done; 1039 g_io_request(cbp, disk->d_consumer); 1040 G_MIRROR_DEBUG(1, "Kernel dump will go to %s.", 1041 g_mirror_get_diskname(disk)); 1042} 1043 1044static void 1045g_mirror_flush(struct g_mirror_softc *sc, struct bio *bp) 1046{ 1047 struct bio_queue_head queue; 1048 struct g_mirror_disk *disk; 1049 struct g_consumer *cp; 1050 struct bio *cbp; 1051 1052 bioq_init(&queue); 1053 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1054 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1055 continue; 1056 cbp = g_clone_bio(bp); 1057 if (cbp == NULL) { 1058 for (cbp = bioq_first(&queue); cbp != NULL; 1059 cbp = bioq_first(&queue)) { 1060 bioq_remove(&queue, cbp); 1061 g_destroy_bio(cbp); 1062 } 1063 if (bp->bio_error == 0) 1064 bp->bio_error = ENOMEM; 1065 g_io_deliver(bp, bp->bio_error); 1066 return; 1067 } 1068 bioq_insert_tail(&queue, cbp); 1069 cbp->bio_done = g_std_done; 1070 cbp->bio_caller1 = disk; 1071 cbp->bio_to = disk->d_consumer->provider; 1072 } 1073 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1074 bioq_remove(&queue, cbp); 1075 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1076 disk = cbp->bio_caller1; 1077 cbp->bio_caller1 = NULL; 1078 cp = disk->d_consumer; 1079 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1080 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1081 cp->acr, cp->acw, cp->ace)); 1082 g_io_request(cbp, disk->d_consumer); 1083 } 1084} 1085 1086static void 1087g_mirror_start(struct bio *bp) 1088{ 1089 struct g_mirror_softc *sc; 1090 1091 sc = bp->bio_to->geom->softc; 1092 /* 1093 * If sc == NULL or there are no valid disks, provider's error 1094 * should be set and g_mirror_start() should not be called at all. 1095 */ 1096 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1097 ("Provider's error should be set (error=%d)(mirror=%s).", 1098 bp->bio_to->error, bp->bio_to->name)); 1099 G_MIRROR_LOGREQ(3, bp, "Request received."); 1100 1101 switch (bp->bio_cmd) { 1102 case BIO_READ: 1103 case BIO_WRITE: 1104 case BIO_DELETE: 1105 break; 1106 case BIO_FLUSH: 1107 g_mirror_flush(sc, bp); 1108 return; 1109 case BIO_GETATTR: 1110 if (strcmp("GEOM::kerneldump", bp->bio_attribute) == 0) { 1111 g_mirror_kernel_dump(bp); 1112 return; 1113 } 1114 /* FALLTHROUGH */ 1115 default: 1116 g_io_deliver(bp, EOPNOTSUPP); 1117 return; 1118 } 1119 mtx_lock(&sc->sc_queue_mtx); 1120 bioq_disksort(&sc->sc_queue, bp); 1121 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1122 wakeup(sc); 1123 mtx_unlock(&sc->sc_queue_mtx); 1124} 1125 1126/* 1127 * Return TRUE if the given request is colliding with a in-progress 1128 * synchronization request. 1129 */ 1130static int 1131g_mirror_sync_collision(struct g_mirror_softc *sc, struct bio *bp) 1132{ 1133 struct g_mirror_disk *disk; 1134 struct bio *sbp; 1135 off_t rstart, rend, sstart, send; 1136 int i; 1137 1138 if (sc->sc_sync.ds_ndisks == 0) 1139 return (0); 1140 rstart = bp->bio_offset; 1141 rend = bp->bio_offset + bp->bio_length; 1142 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1143 if (disk->d_state != G_MIRROR_DISK_STATE_SYNCHRONIZING) 1144 continue; 1145 for (i = 0; i < g_mirror_syncreqs; i++) { 1146 sbp = disk->d_sync.ds_bios[i]; 1147 if (sbp == NULL) 1148 continue; 1149 sstart = sbp->bio_offset; 1150 send = sbp->bio_offset + sbp->bio_length; 1151 if (rend > sstart && rstart < send) 1152 return (1); 1153 } 1154 } 1155 return (0); 1156} 1157 1158/* 1159 * Return TRUE if the given sync request is colliding with a in-progress regular 1160 * request. 1161 */ 1162static int 1163g_mirror_regular_collision(struct g_mirror_softc *sc, struct bio *sbp) 1164{ 1165 off_t rstart, rend, sstart, send; 1166 struct bio *bp; 1167 1168 if (sc->sc_sync.ds_ndisks == 0) 1169 return (0); 1170 sstart = sbp->bio_offset; 1171 send = sbp->bio_offset + sbp->bio_length; 1172 TAILQ_FOREACH(bp, &sc->sc_inflight.queue, bio_queue) { 1173 rstart = bp->bio_offset; 1174 rend = bp->bio_offset + bp->bio_length; 1175 if (rend > sstart && rstart < send) 1176 return (1); 1177 } 1178 return (0); 1179} 1180 1181/* 1182 * Puts request onto delayed queue. 1183 */ 1184static void 1185g_mirror_regular_delay(struct g_mirror_softc *sc, struct bio *bp) 1186{ 1187 1188 G_MIRROR_LOGREQ(2, bp, "Delaying request."); 1189 bioq_insert_head(&sc->sc_regular_delayed, bp); 1190} 1191 1192/* 1193 * Puts synchronization request onto delayed queue. 1194 */ 1195static void 1196g_mirror_sync_delay(struct g_mirror_softc *sc, struct bio *bp) 1197{ 1198 1199 G_MIRROR_LOGREQ(2, bp, "Delaying synchronization request."); 1200 bioq_insert_tail(&sc->sc_sync_delayed, bp); 1201} 1202 1203/* 1204 * Releases delayed regular requests which don't collide anymore with sync 1205 * requests. 1206 */ 1207static void 1208g_mirror_regular_release(struct g_mirror_softc *sc) 1209{ 1210 struct bio *bp, *bp2; 1211 1212 TAILQ_FOREACH_SAFE(bp, &sc->sc_regular_delayed.queue, bio_queue, bp2) { 1213 if (g_mirror_sync_collision(sc, bp)) 1214 continue; 1215 bioq_remove(&sc->sc_regular_delayed, bp); 1216 G_MIRROR_LOGREQ(2, bp, "Releasing delayed request (%p).", bp); 1217 mtx_lock(&sc->sc_queue_mtx); 1218 bioq_insert_head(&sc->sc_queue, bp); 1219#if 0 1220 /* 1221 * wakeup() is not needed, because this function is called from 1222 * the worker thread. 1223 */ 1224 wakeup(&sc->sc_queue); 1225#endif 1226 mtx_unlock(&sc->sc_queue_mtx); 1227 } 1228} 1229 1230/* 1231 * Releases delayed sync requests which don't collide anymore with regular 1232 * requests. 1233 */ 1234static void 1235g_mirror_sync_release(struct g_mirror_softc *sc) 1236{ 1237 struct bio *bp, *bp2; 1238 1239 TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed.queue, bio_queue, bp2) { 1240 if (g_mirror_regular_collision(sc, bp)) 1241 continue; 1242 bioq_remove(&sc->sc_sync_delayed, bp); 1243 G_MIRROR_LOGREQ(2, bp, 1244 "Releasing delayed synchronization request."); 1245 g_io_request(bp, bp->bio_from); 1246 } 1247} 1248 1249/* 1250 * Handle synchronization requests. 1251 * Every synchronization request is two-steps process: first, READ request is 1252 * send to active provider and then WRITE request (with read data) to the provider 1253 * beeing synchronized. When WRITE is finished, new synchronization request is 1254 * send. 1255 */ 1256static void 1257g_mirror_sync_request(struct bio *bp) 1258{ 1259 struct g_mirror_softc *sc; 1260 struct g_mirror_disk *disk; 1261 1262 bp->bio_from->index--; 1263 sc = bp->bio_from->geom->softc; 1264 disk = bp->bio_from->private; 1265 if (disk == NULL) { 1266 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 1267 g_topology_lock(); 1268 g_mirror_kill_consumer(sc, bp->bio_from); 1269 g_topology_unlock(); 1270 free(bp->bio_data, M_MIRROR); 1271 g_destroy_bio(bp); 1272 sx_xlock(&sc->sc_lock); 1273 return; 1274 } 1275 1276 /* 1277 * Synchronization request. 1278 */ 1279 switch (bp->bio_cmd) { 1280 case BIO_READ: 1281 { 1282 struct g_consumer *cp; 1283 1284 if (bp->bio_error != 0) { 1285 G_MIRROR_LOGREQ(0, bp, 1286 "Synchronization request failed (error=%d).", 1287 bp->bio_error); 1288 g_destroy_bio(bp); 1289 return; 1290 } 1291 G_MIRROR_LOGREQ(3, bp, 1292 "Synchronization request half-finished."); 1293 bp->bio_cmd = BIO_WRITE; 1294 bp->bio_cflags = 0; 1295 cp = disk->d_consumer; 1296 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1297 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1298 cp->acr, cp->acw, cp->ace)); 1299 cp->index++; 1300 g_io_request(bp, cp); 1301 return; 1302 } 1303 case BIO_WRITE: 1304 { 1305 struct g_mirror_disk_sync *sync; 1306 off_t offset; 1307 void *data; 1308 int i; 1309 1310 if (bp->bio_error != 0) { 1311 G_MIRROR_LOGREQ(0, bp, 1312 "Synchronization request failed (error=%d).", 1313 bp->bio_error); 1314 g_destroy_bio(bp); 1315 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 1316 g_mirror_event_send(disk, 1317 G_MIRROR_DISK_STATE_DISCONNECTED, 1318 G_MIRROR_EVENT_DONTWAIT); 1319 return; 1320 } 1321 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1322 sync = &disk->d_sync; 1323 if (sync->ds_offset == sc->sc_mediasize || 1324 sync->ds_consumer == NULL || 1325 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1326 /* Don't send more synchronization requests. */ 1327 sync->ds_inflight--; 1328 if (sync->ds_bios != NULL) { 1329 i = (int)(uintptr_t)bp->bio_caller1; 1330 sync->ds_bios[i] = NULL; 1331 } 1332 free(bp->bio_data, M_MIRROR); 1333 g_destroy_bio(bp); 1334 if (sync->ds_inflight > 0) 1335 return; 1336 if (sync->ds_consumer == NULL || 1337 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1338 return; 1339 } 1340 /* Disk up-to-date, activate it. */ 1341 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1342 G_MIRROR_EVENT_DONTWAIT); 1343 return; 1344 } 1345 1346 /* Send next synchronization request. */ 1347 data = bp->bio_data; 1348 bzero(bp, sizeof(*bp)); 1349 bp->bio_cmd = BIO_READ; 1350 bp->bio_offset = sync->ds_offset; 1351 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1352 sync->ds_offset += bp->bio_length; 1353 bp->bio_done = g_mirror_sync_done; 1354 bp->bio_data = data; 1355 bp->bio_from = sync->ds_consumer; 1356 bp->bio_to = sc->sc_provider; 1357 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1358 sync->ds_consumer->index++; 1359 /* 1360 * Delay the request if it is colliding with a regular request. 1361 */ 1362 if (g_mirror_regular_collision(sc, bp)) 1363 g_mirror_sync_delay(sc, bp); 1364 else 1365 g_io_request(bp, sync->ds_consumer); 1366 1367 /* Release delayed requests if possible. */ 1368 g_mirror_regular_release(sc); 1369 1370 /* Find the smallest offset */ 1371 offset = sc->sc_mediasize; 1372 for (i = 0; i < g_mirror_syncreqs; i++) { 1373 bp = sync->ds_bios[i]; 1374 if (bp->bio_offset < offset) 1375 offset = bp->bio_offset; 1376 } 1377 if (sync->ds_offset_done + (MAXPHYS * 100) < offset) { 1378 /* Update offset_done on every 100 blocks. */ 1379 sync->ds_offset_done = offset; 1380 g_mirror_update_metadata(disk); 1381 } 1382 return; 1383 } 1384 default: 1385 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1386 bp->bio_cmd, sc->sc_name)); 1387 break; 1388 } 1389} 1390 1391static void 1392g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1393{ 1394 struct g_mirror_disk *disk; 1395 struct g_consumer *cp; 1396 struct bio *cbp; 1397 1398 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1399 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1400 break; 1401 } 1402 if (disk == NULL) { 1403 if (bp->bio_error == 0) 1404 bp->bio_error = ENXIO; 1405 g_io_deliver(bp, bp->bio_error); 1406 return; 1407 } 1408 cbp = g_clone_bio(bp); 1409 if (cbp == NULL) { 1410 if (bp->bio_error == 0) 1411 bp->bio_error = ENOMEM; 1412 g_io_deliver(bp, bp->bio_error); 1413 return; 1414 } 1415 /* 1416 * Fill in the component buf structure. 1417 */ 1418 cp = disk->d_consumer; 1419 cbp->bio_done = g_mirror_done; 1420 cbp->bio_to = cp->provider; 1421 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1422 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1423 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1424 cp->acw, cp->ace)); 1425 cp->index++; 1426 g_io_request(cbp, cp); 1427} 1428 1429static void 1430g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1431{ 1432 struct g_mirror_disk *disk; 1433 struct g_consumer *cp; 1434 struct bio *cbp; 1435 1436 disk = g_mirror_get_disk(sc); 1437 if (disk == NULL) { 1438 if (bp->bio_error == 0) 1439 bp->bio_error = ENXIO; 1440 g_io_deliver(bp, bp->bio_error); 1441 return; 1442 } 1443 cbp = g_clone_bio(bp); 1444 if (cbp == NULL) { 1445 if (bp->bio_error == 0) 1446 bp->bio_error = ENOMEM; 1447 g_io_deliver(bp, bp->bio_error); 1448 return; 1449 } 1450 /* 1451 * Fill in the component buf structure. 1452 */ 1453 cp = disk->d_consumer; 1454 cbp->bio_done = g_mirror_done; 1455 cbp->bio_to = cp->provider; 1456 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1457 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1458 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1459 cp->acw, cp->ace)); 1460 cp->index++; 1461 g_io_request(cbp, cp); 1462} 1463 1464static void 1465g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1466{ 1467 struct g_mirror_disk *disk, *dp; 1468 struct g_consumer *cp; 1469 struct bio *cbp; 1470 struct bintime curtime; 1471 1472 binuptime(&curtime); 1473 /* 1474 * Find a disk which the smallest load. 1475 */ 1476 disk = NULL; 1477 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1478 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1479 continue; 1480 /* If disk wasn't used for more than 2 sec, use it. */ 1481 if (curtime.sec - dp->d_last_used.sec >= 2) { 1482 disk = dp; 1483 break; 1484 } 1485 if (disk == NULL || 1486 bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) { 1487 disk = dp; 1488 } 1489 } 1490 KASSERT(disk != NULL, ("NULL disk for %s.", sc->sc_name)); 1491 cbp = g_clone_bio(bp); 1492 if (cbp == NULL) { 1493 if (bp->bio_error == 0) 1494 bp->bio_error = ENOMEM; 1495 g_io_deliver(bp, bp->bio_error); 1496 return; 1497 } 1498 /* 1499 * Fill in the component buf structure. 1500 */ 1501 cp = disk->d_consumer; 1502 cbp->bio_done = g_mirror_done; 1503 cbp->bio_to = cp->provider; 1504 binuptime(&disk->d_last_used); 1505 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1506 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1507 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1508 cp->acw, cp->ace)); 1509 cp->index++; 1510 g_io_request(cbp, cp); 1511} 1512 1513static void 1514g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1515{ 1516 struct bio_queue_head queue; 1517 struct g_mirror_disk *disk; 1518 struct g_consumer *cp; 1519 struct bio *cbp; 1520 off_t left, mod, offset, slice; 1521 u_char *data; 1522 u_int ndisks; 1523 1524 if (bp->bio_length <= sc->sc_slice) { 1525 g_mirror_request_round_robin(sc, bp); 1526 return; 1527 } 1528 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1529 slice = bp->bio_length / ndisks; 1530 mod = slice % sc->sc_provider->sectorsize; 1531 if (mod != 0) 1532 slice += sc->sc_provider->sectorsize - mod; 1533 /* 1534 * Allocate all bios before sending any request, so we can 1535 * return ENOMEM in nice and clean way. 1536 */ 1537 left = bp->bio_length; 1538 offset = bp->bio_offset; 1539 data = bp->bio_data; 1540 bioq_init(&queue); 1541 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1542 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1543 continue; 1544 cbp = g_clone_bio(bp); 1545 if (cbp == NULL) { 1546 for (cbp = bioq_first(&queue); cbp != NULL; 1547 cbp = bioq_first(&queue)) { 1548 bioq_remove(&queue, cbp); 1549 g_destroy_bio(cbp); 1550 } 1551 if (bp->bio_error == 0) 1552 bp->bio_error = ENOMEM; 1553 g_io_deliver(bp, bp->bio_error); 1554 return; 1555 } 1556 bioq_insert_tail(&queue, cbp); 1557 cbp->bio_done = g_mirror_done; 1558 cbp->bio_caller1 = disk; 1559 cbp->bio_to = disk->d_consumer->provider; 1560 cbp->bio_offset = offset; 1561 cbp->bio_data = data; 1562 cbp->bio_length = MIN(left, slice); 1563 left -= cbp->bio_length; 1564 if (left == 0) 1565 break; 1566 offset += cbp->bio_length; 1567 data += cbp->bio_length; 1568 } 1569 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1570 bioq_remove(&queue, cbp); 1571 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1572 disk = cbp->bio_caller1; 1573 cbp->bio_caller1 = NULL; 1574 cp = disk->d_consumer; 1575 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1576 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1577 cp->acr, cp->acw, cp->ace)); 1578 disk->d_consumer->index++; 1579 g_io_request(cbp, disk->d_consumer); 1580 } 1581} 1582 1583static void 1584g_mirror_register_request(struct bio *bp) 1585{ 1586 struct g_mirror_softc *sc; 1587 1588 sc = bp->bio_to->geom->softc; 1589 switch (bp->bio_cmd) { 1590 case BIO_READ: 1591 switch (sc->sc_balance) { 1592 case G_MIRROR_BALANCE_LOAD: 1593 g_mirror_request_load(sc, bp); 1594 break; 1595 case G_MIRROR_BALANCE_PREFER: 1596 g_mirror_request_prefer(sc, bp); 1597 break; 1598 case G_MIRROR_BALANCE_ROUND_ROBIN: 1599 g_mirror_request_round_robin(sc, bp); 1600 break; 1601 case G_MIRROR_BALANCE_SPLIT: 1602 g_mirror_request_split(sc, bp); 1603 break; 1604 } 1605 return; 1606 case BIO_WRITE: 1607 case BIO_DELETE: 1608 { 1609 struct g_mirror_disk *disk; 1610 struct g_mirror_disk_sync *sync; 1611 struct bio_queue_head queue; 1612 struct g_consumer *cp; 1613 struct bio *cbp; 1614 1615 /* 1616 * Delay the request if it is colliding with a synchronization 1617 * request. 1618 */ 1619 if (g_mirror_sync_collision(sc, bp)) { 1620 g_mirror_regular_delay(sc, bp); 1621 return; 1622 } 1623 1624 if (sc->sc_idle) 1625 g_mirror_unidle(sc); 1626 else 1627 sc->sc_last_write = time_uptime; 1628 1629 /* 1630 * Allocate all bios before sending any request, so we can 1631 * return ENOMEM in nice and clean way. 1632 */ 1633 bioq_init(&queue); 1634 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1635 sync = &disk->d_sync; 1636 switch (disk->d_state) { 1637 case G_MIRROR_DISK_STATE_ACTIVE: 1638 break; 1639 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1640 if (bp->bio_offset >= sync->ds_offset) 1641 continue; 1642 break; 1643 default: 1644 continue; 1645 } 1646 cbp = g_clone_bio(bp); 1647 if (cbp == NULL) { 1648 for (cbp = bioq_first(&queue); cbp != NULL; 1649 cbp = bioq_first(&queue)) { 1650 bioq_remove(&queue, cbp); 1651 g_destroy_bio(cbp); 1652 } 1653 if (bp->bio_error == 0) 1654 bp->bio_error = ENOMEM; 1655 g_io_deliver(bp, bp->bio_error); 1656 return; 1657 } 1658 bioq_insert_tail(&queue, cbp); 1659 cbp->bio_done = g_mirror_done; 1660 cp = disk->d_consumer; 1661 cbp->bio_caller1 = cp; 1662 cbp->bio_to = cp->provider; 1663 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1664 ("Consumer %s not opened (r%dw%de%d).", 1665 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1666 } 1667 for (cbp = bioq_first(&queue); cbp != NULL; 1668 cbp = bioq_first(&queue)) { 1669 bioq_remove(&queue, cbp); 1670 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1671 cp = cbp->bio_caller1; 1672 cbp->bio_caller1 = NULL; 1673 cp->index++; 1674 sc->sc_writes++; 1675 g_io_request(cbp, cp); 1676 } 1677 /* 1678 * Put request onto inflight queue, so we can check if new 1679 * synchronization requests don't collide with it. 1680 */ 1681 bioq_insert_tail(&sc->sc_inflight, bp); 1682 /* 1683 * Bump syncid on first write. 1684 */ 1685 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1686 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1687 g_mirror_bump_syncid(sc); 1688 } 1689 return; 1690 } 1691 default: 1692 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1693 bp->bio_cmd, sc->sc_name)); 1694 break; 1695 } 1696} 1697 1698static int 1699g_mirror_can_destroy(struct g_mirror_softc *sc) 1700{ 1701 struct g_geom *gp; 1702 struct g_consumer *cp; 1703 1704 g_topology_assert(); 1705 gp = sc->sc_geom; 1706 if (gp->softc == NULL) 1707 return (1); 1708 LIST_FOREACH(cp, &gp->consumer, consumer) { 1709 if (g_mirror_is_busy(sc, cp)) 1710 return (0); 1711 } 1712 gp = sc->sc_sync.ds_geom; 1713 LIST_FOREACH(cp, &gp->consumer, consumer) { 1714 if (g_mirror_is_busy(sc, cp)) 1715 return (0); 1716 } 1717 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1718 sc->sc_name); 1719 return (1); 1720} 1721 1722static int 1723g_mirror_try_destroy(struct g_mirror_softc *sc) 1724{ 1725 1726 if (sc->sc_rootmount != NULL) { 1727 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1728 sc->sc_rootmount); 1729 root_mount_rel(sc->sc_rootmount); 1730 sc->sc_rootmount = NULL; 1731 } 1732 g_topology_lock(); 1733 if (!g_mirror_can_destroy(sc)) { 1734 g_topology_unlock(); 1735 return (0); 1736 } 1737 sc->sc_geom->softc = NULL; 1738 sc->sc_sync.ds_geom->softc = NULL; 1739 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1740 g_topology_unlock(); 1741 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1742 &sc->sc_worker); 1743 /* Unlock sc_lock here, as it can be destroyed after wakeup. */ 1744 sx_xunlock(&sc->sc_lock); 1745 wakeup(&sc->sc_worker); 1746 sc->sc_worker = NULL; 1747 } else { 1748 g_topology_unlock(); 1749 g_mirror_destroy_device(sc); 1750 free(sc, M_MIRROR); 1751 } 1752 return (1); 1753} 1754 1755/* 1756 * Worker thread. 1757 */ 1758static void 1759g_mirror_worker(void *arg) 1760{ 1761 struct g_mirror_softc *sc; 1762 struct g_mirror_event *ep; 1763 struct bio *bp; 1764 int timeout; 1765 1766 sc = arg; 1767 mtx_lock_spin(&sched_lock); 1768 sched_prio(curthread, PRIBIO); 1769 mtx_unlock_spin(&sched_lock); 1770 1771 sx_xlock(&sc->sc_lock); 1772 for (;;) { 1773 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1774 /* 1775 * First take a look at events. 1776 * This is important to handle events before any I/O requests. 1777 */ 1778 ep = g_mirror_event_get(sc); 1779 if (ep != NULL) { 1780 g_mirror_event_remove(sc, ep); 1781 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1782 /* Update only device status. */ 1783 G_MIRROR_DEBUG(3, 1784 "Running event for device %s.", 1785 sc->sc_name); 1786 ep->e_error = 0; 1787 g_mirror_update_device(sc, 1); 1788 } else { 1789 /* Update disk status. */ 1790 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1791 g_mirror_get_diskname(ep->e_disk)); 1792 ep->e_error = g_mirror_update_disk(ep->e_disk, 1793 ep->e_state); 1794 if (ep->e_error == 0) 1795 g_mirror_update_device(sc, 0); 1796 } 1797 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1798 KASSERT(ep->e_error == 0, 1799 ("Error cannot be handled.")); 1800 g_mirror_event_free(ep); 1801 } else { 1802 ep->e_flags |= G_MIRROR_EVENT_DONE; 1803 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1804 ep); 1805 mtx_lock(&sc->sc_events_mtx); 1806 wakeup(ep); 1807 mtx_unlock(&sc->sc_events_mtx); 1808 } 1809 if ((sc->sc_flags & 1810 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1811 if (g_mirror_try_destroy(sc)) { 1812 curthread->td_pflags &= ~TDP_GEOM; 1813 G_MIRROR_DEBUG(1, "Thread exiting."); 1814 kthread_exit(0); 1815 } 1816 } 1817 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1818 continue; 1819 } 1820 /* 1821 * Check if we can mark array as CLEAN and if we can't take 1822 * how much seconds should we wait. 1823 */ 1824 timeout = g_mirror_idle(sc, -1); 1825 /* 1826 * Now I/O requests. 1827 */ 1828 /* Get first request from the queue. */ 1829 mtx_lock(&sc->sc_queue_mtx); 1830 bp = bioq_first(&sc->sc_queue); 1831 if (bp == NULL) { 1832 if ((sc->sc_flags & 1833 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1834 mtx_unlock(&sc->sc_queue_mtx); 1835 if (g_mirror_try_destroy(sc)) { 1836 curthread->td_pflags &= ~TDP_GEOM; 1837 G_MIRROR_DEBUG(1, "Thread exiting."); 1838 kthread_exit(0); 1839 } 1840 mtx_lock(&sc->sc_queue_mtx); 1841 } 1842 sx_xunlock(&sc->sc_lock); 1843 /* 1844 * XXX: We can miss an event here, because an event 1845 * can be added without sx-device-lock and without 1846 * mtx-queue-lock. Maybe I should just stop using 1847 * dedicated mutex for events synchronization and 1848 * stick with the queue lock? 1849 * The event will hang here until next I/O request 1850 * or next event is received. 1851 */ 1852 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 1853 timeout * hz); 1854 sx_xlock(&sc->sc_lock); 1855 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1856 continue; 1857 } 1858 bioq_remove(&sc->sc_queue, bp); 1859 mtx_unlock(&sc->sc_queue_mtx); 1860 1861 if (bp->bio_from->geom == sc->sc_sync.ds_geom && 1862 (bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1863 g_mirror_sync_request(bp); /* READ */ 1864 } else if (bp->bio_to != sc->sc_provider) { 1865 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) 1866 g_mirror_regular_request(bp); 1867 else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 1868 g_mirror_sync_request(bp); /* WRITE */ 1869 else { 1870 KASSERT(0, 1871 ("Invalid request cflags=0x%hhx to=%s.", 1872 bp->bio_cflags, bp->bio_to->name)); 1873 } 1874 } else { 1875 g_mirror_register_request(bp); 1876 } 1877 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1878 } 1879} 1880 1881static void 1882g_mirror_update_idle(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 1883{ 1884 1885 sx_assert(&sc->sc_lock, SX_LOCKED); 1886 1887 if (!sc->sc_idle && (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1888 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 1889 g_mirror_get_diskname(disk), sc->sc_name); 1890 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1891 } else if (sc->sc_idle && 1892 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1893 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 1894 g_mirror_get_diskname(disk), sc->sc_name); 1895 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1896 } 1897} 1898 1899static void 1900g_mirror_sync_start(struct g_mirror_disk *disk) 1901{ 1902 struct g_mirror_softc *sc; 1903 struct g_consumer *cp; 1904 struct bio *bp; 1905 int error, i; 1906 1907 g_topology_assert_not(); 1908 sc = disk->d_softc; 1909 sx_assert(&sc->sc_lock, SX_LOCKED); 1910 1911 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1912 ("Disk %s is not marked for synchronization.", 1913 g_mirror_get_diskname(disk))); 1914 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1915 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1916 sc->sc_state)); 1917 1918 sx_xunlock(&sc->sc_lock); 1919 g_topology_lock(); 1920 cp = g_new_consumer(sc->sc_sync.ds_geom); 1921 error = g_attach(cp, sc->sc_provider); 1922 KASSERT(error == 0, 1923 ("Cannot attach to %s (error=%d).", sc->sc_name, error)); 1924 error = g_access(cp, 1, 0, 0); 1925 KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error)); 1926 g_topology_unlock(); 1927 sx_xlock(&sc->sc_lock); 1928 1929 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1930 g_mirror_get_diskname(disk)); 1931 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1932 KASSERT(disk->d_sync.ds_consumer == NULL, 1933 ("Sync consumer already exists (device=%s, disk=%s).", 1934 sc->sc_name, g_mirror_get_diskname(disk))); 1935 1936 disk->d_sync.ds_consumer = cp; 1937 disk->d_sync.ds_consumer->private = disk; 1938 disk->d_sync.ds_consumer->index = 0; 1939 1940 /* 1941 * Allocate memory for synchronization bios and initialize them. 1942 */ 1943 disk->d_sync.ds_bios = malloc(sizeof(struct bio *) * g_mirror_syncreqs, 1944 M_MIRROR, M_WAITOK); 1945 for (i = 0; i < g_mirror_syncreqs; i++) { 1946 bp = g_alloc_bio(); 1947 disk->d_sync.ds_bios[i] = bp; 1948 bp->bio_parent = NULL; 1949 bp->bio_cmd = BIO_READ; 1950 bp->bio_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 1951 bp->bio_cflags = 0; 1952 bp->bio_offset = disk->d_sync.ds_offset; 1953 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1954 disk->d_sync.ds_offset += bp->bio_length; 1955 bp->bio_done = g_mirror_sync_done; 1956 bp->bio_from = disk->d_sync.ds_consumer; 1957 bp->bio_to = sc->sc_provider; 1958 bp->bio_caller1 = (void *)(uintptr_t)i; 1959 } 1960 1961 /* Increase the number of disks in SYNCHRONIZING state. */ 1962 sc->sc_sync.ds_ndisks++; 1963 /* Set the number of in-flight synchronization requests. */ 1964 disk->d_sync.ds_inflight = g_mirror_syncreqs; 1965 1966 /* 1967 * Fire off first synchronization requests. 1968 */ 1969 for (i = 0; i < g_mirror_syncreqs; i++) { 1970 bp = disk->d_sync.ds_bios[i]; 1971 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1972 disk->d_sync.ds_consumer->index++; 1973 /* 1974 * Delay the request if it is colliding with a regular request. 1975 */ 1976 if (g_mirror_regular_collision(sc, bp)) 1977 g_mirror_sync_delay(sc, bp); 1978 else 1979 g_io_request(bp, disk->d_sync.ds_consumer); 1980 } 1981} 1982 1983/* 1984 * Stop synchronization process. 1985 * type: 0 - synchronization finished 1986 * 1 - synchronization stopped 1987 */ 1988static void 1989g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1990{ 1991 struct g_mirror_softc *sc; 1992 struct g_consumer *cp; 1993 1994 g_topology_assert_not(); 1995 sc = disk->d_softc; 1996 sx_assert(&sc->sc_lock, SX_LOCKED); 1997 1998 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1999 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2000 g_mirror_disk_state2str(disk->d_state))); 2001 if (disk->d_sync.ds_consumer == NULL) 2002 return; 2003 2004 if (type == 0) { 2005 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 2006 sc->sc_name, g_mirror_get_diskname(disk)); 2007 } else /* if (type == 1) */ { 2008 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 2009 sc->sc_name, g_mirror_get_diskname(disk)); 2010 } 2011 free(disk->d_sync.ds_bios, M_MIRROR); 2012 disk->d_sync.ds_bios = NULL; 2013 cp = disk->d_sync.ds_consumer; 2014 disk->d_sync.ds_consumer = NULL; 2015 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2016 sc->sc_sync.ds_ndisks--; 2017 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 2018 g_topology_lock(); 2019 g_mirror_kill_consumer(sc, cp); 2020 g_topology_unlock(); 2021 sx_xlock(&sc->sc_lock); 2022} 2023 2024static void 2025g_mirror_launch_provider(struct g_mirror_softc *sc) 2026{ 2027 struct g_mirror_disk *disk; 2028 struct g_provider *pp; 2029 2030 sx_assert(&sc->sc_lock, SX_LOCKED); 2031 2032 g_topology_lock(); 2033 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 2034 pp->mediasize = sc->sc_mediasize; 2035 pp->sectorsize = sc->sc_sectorsize; 2036 sc->sc_provider = pp; 2037 g_error_provider(pp, 0); 2038 g_topology_unlock(); 2039 G_MIRROR_DEBUG(0, "Device %s launched (%u/%u).", pp->name, 2040 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE), sc->sc_ndisks); 2041 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2042 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2043 g_mirror_sync_start(disk); 2044 } 2045} 2046 2047static void 2048g_mirror_destroy_provider(struct g_mirror_softc *sc) 2049{ 2050 struct g_mirror_disk *disk; 2051 struct bio *bp; 2052 2053 g_topology_assert_not(); 2054 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 2055 sc->sc_name)); 2056 2057 g_topology_lock(); 2058 g_error_provider(sc->sc_provider, ENXIO); 2059 mtx_lock(&sc->sc_queue_mtx); 2060 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 2061 bioq_remove(&sc->sc_queue, bp); 2062 g_io_deliver(bp, ENXIO); 2063 } 2064 mtx_unlock(&sc->sc_queue_mtx); 2065 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 2066 sc->sc_provider->name); 2067 sc->sc_provider->flags |= G_PF_WITHER; 2068 g_orphan_provider(sc->sc_provider, ENXIO); 2069 g_topology_unlock(); 2070 sc->sc_provider = NULL; 2071 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2072 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2073 g_mirror_sync_stop(disk, 1); 2074 } 2075} 2076 2077static void 2078g_mirror_go(void *arg) 2079{ 2080 struct g_mirror_softc *sc; 2081 2082 sc = arg; 2083 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 2084 g_mirror_event_send(sc, 0, 2085 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 2086} 2087 2088static u_int 2089g_mirror_determine_state(struct g_mirror_disk *disk) 2090{ 2091 struct g_mirror_softc *sc; 2092 u_int state; 2093 2094 sc = disk->d_softc; 2095 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 2096 if ((disk->d_flags & 2097 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2098 /* Disk does not need synchronization. */ 2099 state = G_MIRROR_DISK_STATE_ACTIVE; 2100 } else { 2101 if ((sc->sc_flags & 2102 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2103 (disk->d_flags & 2104 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2105 /* 2106 * We can start synchronization from 2107 * the stored offset. 2108 */ 2109 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2110 } else { 2111 state = G_MIRROR_DISK_STATE_STALE; 2112 } 2113 } 2114 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 2115 /* 2116 * Reset all synchronization data for this disk, 2117 * because if it even was synchronized, it was 2118 * synchronized to disks with different syncid. 2119 */ 2120 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2121 disk->d_sync.ds_offset = 0; 2122 disk->d_sync.ds_offset_done = 0; 2123 disk->d_sync.ds_syncid = sc->sc_syncid; 2124 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2125 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2126 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2127 } else { 2128 state = G_MIRROR_DISK_STATE_STALE; 2129 } 2130 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 2131 /* 2132 * Not good, NOT GOOD! 2133 * It means that mirror was started on stale disks 2134 * and more fresh disk just arrive. 2135 * If there were writes, mirror is broken, sorry. 2136 * I think the best choice here is don't touch 2137 * this disk and inform the user loudly. 2138 */ 2139 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 2140 "disk (%s) arrives!! It will not be connected to the " 2141 "running device.", sc->sc_name, 2142 g_mirror_get_diskname(disk)); 2143 g_mirror_destroy_disk(disk); 2144 state = G_MIRROR_DISK_STATE_NONE; 2145 /* Return immediately, because disk was destroyed. */ 2146 return (state); 2147 } 2148 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 2149 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 2150 return (state); 2151} 2152 2153/* 2154 * Update device state. 2155 */ 2156static void 2157g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 2158{ 2159 struct g_mirror_disk *disk; 2160 u_int state; 2161 2162 sx_assert(&sc->sc_lock, SX_XLOCKED); 2163 2164 switch (sc->sc_state) { 2165 case G_MIRROR_DEVICE_STATE_STARTING: 2166 { 2167 struct g_mirror_disk *pdisk, *tdisk; 2168 u_int dirty, ndisks, genid, syncid; 2169 2170 KASSERT(sc->sc_provider == NULL, 2171 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 2172 /* 2173 * Are we ready? We are, if all disks are connected or 2174 * if we have any disks and 'force' is true. 2175 */ 2176 ndisks = g_mirror_ndisks(sc, -1); 2177 if (sc->sc_ndisks == ndisks || (force && ndisks > 0)) { 2178 ; 2179 } else if (ndisks == 0) { 2180 /* 2181 * Disks went down in starting phase, so destroy 2182 * device. 2183 */ 2184 callout_drain(&sc->sc_callout); 2185 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2186 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2187 sc->sc_rootmount); 2188 root_mount_rel(sc->sc_rootmount); 2189 sc->sc_rootmount = NULL; 2190 return; 2191 } else { 2192 return; 2193 } 2194 2195 /* 2196 * Activate all disks with the biggest syncid. 2197 */ 2198 if (force) { 2199 /* 2200 * If 'force' is true, we have been called due to 2201 * timeout, so don't bother canceling timeout. 2202 */ 2203 ndisks = 0; 2204 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2205 if ((disk->d_flags & 2206 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2207 ndisks++; 2208 } 2209 } 2210 if (ndisks == 0) { 2211 /* No valid disks found, destroy device. */ 2212 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2213 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2214 __LINE__, sc->sc_rootmount); 2215 root_mount_rel(sc->sc_rootmount); 2216 sc->sc_rootmount = NULL; 2217 return; 2218 } 2219 } else { 2220 /* Cancel timeout. */ 2221 callout_drain(&sc->sc_callout); 2222 } 2223 2224 /* 2225 * Find the biggest genid. 2226 */ 2227 genid = 0; 2228 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2229 if (disk->d_genid > genid) 2230 genid = disk->d_genid; 2231 } 2232 sc->sc_genid = genid; 2233 /* 2234 * Remove all disks without the biggest genid. 2235 */ 2236 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 2237 if (disk->d_genid < genid) { 2238 G_MIRROR_DEBUG(0, 2239 "Component %s (device %s) broken, skipping.", 2240 g_mirror_get_diskname(disk), sc->sc_name); 2241 g_mirror_destroy_disk(disk); 2242 } 2243 } 2244 2245 /* 2246 * Find the biggest syncid. 2247 */ 2248 syncid = 0; 2249 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2250 if (disk->d_sync.ds_syncid > syncid) 2251 syncid = disk->d_sync.ds_syncid; 2252 } 2253 2254 /* 2255 * Here we need to look for dirty disks and if all disks 2256 * with the biggest syncid are dirty, we have to choose 2257 * one with the biggest priority and rebuild the rest. 2258 */ 2259 /* 2260 * Find the number of dirty disks with the biggest syncid. 2261 * Find the number of disks with the biggest syncid. 2262 * While here, find a disk with the biggest priority. 2263 */ 2264 dirty = ndisks = 0; 2265 pdisk = NULL; 2266 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2267 if (disk->d_sync.ds_syncid != syncid) 2268 continue; 2269 if ((disk->d_flags & 2270 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2271 continue; 2272 } 2273 ndisks++; 2274 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2275 dirty++; 2276 if (pdisk == NULL || 2277 pdisk->d_priority < disk->d_priority) { 2278 pdisk = disk; 2279 } 2280 } 2281 } 2282 if (dirty == 0) { 2283 /* No dirty disks at all, great. */ 2284 } else if (dirty == ndisks) { 2285 /* 2286 * Force synchronization for all dirty disks except one 2287 * with the biggest priority. 2288 */ 2289 KASSERT(pdisk != NULL, ("pdisk == NULL")); 2290 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 2291 "master disk for synchronization.", 2292 g_mirror_get_diskname(pdisk), sc->sc_name); 2293 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2294 if (disk->d_sync.ds_syncid != syncid) 2295 continue; 2296 if ((disk->d_flags & 2297 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2298 continue; 2299 } 2300 KASSERT((disk->d_flags & 2301 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2302 ("Disk %s isn't marked as dirty.", 2303 g_mirror_get_diskname(disk))); 2304 /* Skip the disk with the biggest priority. */ 2305 if (disk == pdisk) 2306 continue; 2307 disk->d_sync.ds_syncid = 0; 2308 } 2309 } else if (dirty < ndisks) { 2310 /* 2311 * Force synchronization for all dirty disks. 2312 * We have some non-dirty disks. 2313 */ 2314 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2315 if (disk->d_sync.ds_syncid != syncid) 2316 continue; 2317 if ((disk->d_flags & 2318 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2319 continue; 2320 } 2321 if ((disk->d_flags & 2322 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2323 continue; 2324 } 2325 disk->d_sync.ds_syncid = 0; 2326 } 2327 } 2328 2329 /* Reset hint. */ 2330 sc->sc_hint = NULL; 2331 sc->sc_syncid = syncid; 2332 if (force) { 2333 /* Remember to bump syncid on first write. */ 2334 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2335 } 2336 state = G_MIRROR_DEVICE_STATE_RUNNING; 2337 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2338 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2339 g_mirror_device_state2str(state)); 2340 sc->sc_state = state; 2341 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2342 state = g_mirror_determine_state(disk); 2343 g_mirror_event_send(disk, state, 2344 G_MIRROR_EVENT_DONTWAIT); 2345 if (state == G_MIRROR_DISK_STATE_STALE) 2346 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2347 } 2348 break; 2349 } 2350 case G_MIRROR_DEVICE_STATE_RUNNING: 2351 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2352 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2353 /* 2354 * No active disks or no disks at all, 2355 * so destroy device. 2356 */ 2357 if (sc->sc_provider != NULL) 2358 g_mirror_destroy_provider(sc); 2359 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2360 break; 2361 } else if (g_mirror_ndisks(sc, 2362 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2363 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2364 /* 2365 * We have active disks, launch provider if it doesn't 2366 * exist. 2367 */ 2368 if (sc->sc_provider == NULL) 2369 g_mirror_launch_provider(sc); 2370 if (sc->sc_rootmount != NULL) { 2371 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2372 __LINE__, sc->sc_rootmount); 2373 root_mount_rel(sc->sc_rootmount); 2374 sc->sc_rootmount = NULL; 2375 } 2376 } 2377 /* 2378 * Genid should be bumped immediately, so do it here. 2379 */ 2380 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2381 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2382 g_mirror_bump_genid(sc); 2383 } 2384 break; 2385 default: 2386 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2387 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2388 break; 2389 } 2390} 2391 2392/* 2393 * Update disk state and device state if needed. 2394 */ 2395#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2396 "Disk %s state changed from %s to %s (device %s).", \ 2397 g_mirror_get_diskname(disk), \ 2398 g_mirror_disk_state2str(disk->d_state), \ 2399 g_mirror_disk_state2str(state), sc->sc_name) 2400static int 2401g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2402{ 2403 struct g_mirror_softc *sc; 2404 2405 sc = disk->d_softc; 2406 sx_assert(&sc->sc_lock, SX_XLOCKED); 2407 2408again: 2409 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2410 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2411 g_mirror_disk_state2str(state)); 2412 switch (state) { 2413 case G_MIRROR_DISK_STATE_NEW: 2414 /* 2415 * Possible scenarios: 2416 * 1. New disk arrive. 2417 */ 2418 /* Previous state should be NONE. */ 2419 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2420 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2421 g_mirror_disk_state2str(disk->d_state))); 2422 DISK_STATE_CHANGED(); 2423 2424 disk->d_state = state; 2425 if (LIST_EMPTY(&sc->sc_disks)) 2426 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2427 else { 2428 struct g_mirror_disk *dp; 2429 2430 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2431 if (disk->d_priority >= dp->d_priority) { 2432 LIST_INSERT_BEFORE(dp, disk, d_next); 2433 dp = NULL; 2434 break; 2435 } 2436 if (LIST_NEXT(dp, d_next) == NULL) 2437 break; 2438 } 2439 if (dp != NULL) 2440 LIST_INSERT_AFTER(dp, disk, d_next); 2441 } 2442 G_MIRROR_DEBUG(1, "Device %s: provider %s detected.", 2443 sc->sc_name, g_mirror_get_diskname(disk)); 2444 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2445 break; 2446 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2447 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2448 g_mirror_device_state2str(sc->sc_state), 2449 g_mirror_get_diskname(disk), 2450 g_mirror_disk_state2str(disk->d_state))); 2451 state = g_mirror_determine_state(disk); 2452 if (state != G_MIRROR_DISK_STATE_NONE) 2453 goto again; 2454 break; 2455 case G_MIRROR_DISK_STATE_ACTIVE: 2456 /* 2457 * Possible scenarios: 2458 * 1. New disk does not need synchronization. 2459 * 2. Synchronization process finished successfully. 2460 */ 2461 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2462 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2463 g_mirror_device_state2str(sc->sc_state), 2464 g_mirror_get_diskname(disk), 2465 g_mirror_disk_state2str(disk->d_state))); 2466 /* Previous state should be NEW or SYNCHRONIZING. */ 2467 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2468 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2469 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2470 g_mirror_disk_state2str(disk->d_state))); 2471 DISK_STATE_CHANGED(); 2472 2473 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2474 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2475 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2476 g_mirror_sync_stop(disk, 0); 2477 } 2478 disk->d_state = state; 2479 disk->d_sync.ds_offset = 0; 2480 disk->d_sync.ds_offset_done = 0; 2481 g_mirror_update_idle(sc, disk); 2482 g_mirror_update_metadata(disk); 2483 G_MIRROR_DEBUG(1, "Device %s: provider %s activated.", 2484 sc->sc_name, g_mirror_get_diskname(disk)); 2485 break; 2486 case G_MIRROR_DISK_STATE_STALE: 2487 /* 2488 * Possible scenarios: 2489 * 1. Stale disk was connected. 2490 */ 2491 /* Previous state should be NEW. */ 2492 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2493 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2494 g_mirror_disk_state2str(disk->d_state))); 2495 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2496 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2497 g_mirror_device_state2str(sc->sc_state), 2498 g_mirror_get_diskname(disk), 2499 g_mirror_disk_state2str(disk->d_state))); 2500 /* 2501 * STALE state is only possible if device is marked 2502 * NOAUTOSYNC. 2503 */ 2504 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2505 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2506 g_mirror_device_state2str(sc->sc_state), 2507 g_mirror_get_diskname(disk), 2508 g_mirror_disk_state2str(disk->d_state))); 2509 DISK_STATE_CHANGED(); 2510 2511 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2512 disk->d_state = state; 2513 g_mirror_update_metadata(disk); 2514 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2515 sc->sc_name, g_mirror_get_diskname(disk)); 2516 break; 2517 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2518 /* 2519 * Possible scenarios: 2520 * 1. Disk which needs synchronization was connected. 2521 */ 2522 /* Previous state should be NEW. */ 2523 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2524 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2525 g_mirror_disk_state2str(disk->d_state))); 2526 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2527 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2528 g_mirror_device_state2str(sc->sc_state), 2529 g_mirror_get_diskname(disk), 2530 g_mirror_disk_state2str(disk->d_state))); 2531 DISK_STATE_CHANGED(); 2532 2533 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2534 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2535 disk->d_state = state; 2536 if (sc->sc_provider != NULL) { 2537 g_mirror_sync_start(disk); 2538 g_mirror_update_metadata(disk); 2539 } 2540 break; 2541 case G_MIRROR_DISK_STATE_DISCONNECTED: 2542 /* 2543 * Possible scenarios: 2544 * 1. Device wasn't running yet, but disk disappear. 2545 * 2. Disk was active and disapppear. 2546 * 3. Disk disappear during synchronization process. 2547 */ 2548 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2549 /* 2550 * Previous state should be ACTIVE, STALE or 2551 * SYNCHRONIZING. 2552 */ 2553 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2554 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2555 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2556 ("Wrong disk state (%s, %s).", 2557 g_mirror_get_diskname(disk), 2558 g_mirror_disk_state2str(disk->d_state))); 2559 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2560 /* Previous state should be NEW. */ 2561 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2562 ("Wrong disk state (%s, %s).", 2563 g_mirror_get_diskname(disk), 2564 g_mirror_disk_state2str(disk->d_state))); 2565 /* 2566 * Reset bumping syncid if disk disappeared in STARTING 2567 * state. 2568 */ 2569 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2570 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2571#ifdef INVARIANTS 2572 } else { 2573 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2574 sc->sc_name, 2575 g_mirror_device_state2str(sc->sc_state), 2576 g_mirror_get_diskname(disk), 2577 g_mirror_disk_state2str(disk->d_state))); 2578#endif 2579 } 2580 DISK_STATE_CHANGED(); 2581 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2582 sc->sc_name, g_mirror_get_diskname(disk)); 2583 2584 g_mirror_destroy_disk(disk); 2585 break; 2586 case G_MIRROR_DISK_STATE_DESTROY: 2587 { 2588 int error; 2589 2590 error = g_mirror_clear_metadata(disk); 2591 if (error != 0) 2592 return (error); 2593 DISK_STATE_CHANGED(); 2594 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2595 sc->sc_name, g_mirror_get_diskname(disk)); 2596 2597 g_mirror_destroy_disk(disk); 2598 sc->sc_ndisks--; 2599 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2600 g_mirror_update_metadata(disk); 2601 } 2602 break; 2603 } 2604 default: 2605 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2606 break; 2607 } 2608 return (0); 2609} 2610#undef DISK_STATE_CHANGED 2611 2612int 2613g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2614{ 2615 struct g_provider *pp; 2616 u_char *buf; 2617 int error; 2618 2619 g_topology_assert(); 2620 2621 error = g_access(cp, 1, 0, 0); 2622 if (error != 0) 2623 return (error); 2624 pp = cp->provider; 2625 g_topology_unlock(); 2626 /* Metadata are stored on last sector. */ 2627 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2628 &error); 2629 g_topology_lock(); 2630 g_access(cp, -1, 0, 0); 2631 if (buf == NULL) { 2632 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2633 cp->provider->name, error); 2634 return (error); 2635 } 2636 2637 /* Decode metadata. */ 2638 error = mirror_metadata_decode(buf, md); 2639 g_free(buf); 2640 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2641 return (EINVAL); 2642 if (md->md_version > G_MIRROR_VERSION) { 2643 G_MIRROR_DEBUG(0, 2644 "Kernel module is too old to handle metadata from %s.", 2645 cp->provider->name); 2646 return (EINVAL); 2647 } 2648 if (error != 0) { 2649 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2650 cp->provider->name); 2651 return (error); 2652 } 2653 2654 return (0); 2655} 2656 2657static int 2658g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2659 struct g_mirror_metadata *md) 2660{ 2661 2662 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2663 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2664 pp->name, md->md_did); 2665 return (EEXIST); 2666 } 2667 if (md->md_all != sc->sc_ndisks) { 2668 G_MIRROR_DEBUG(1, 2669 "Invalid '%s' field on disk %s (device %s), skipping.", 2670 "md_all", pp->name, sc->sc_name); 2671 return (EINVAL); 2672 } 2673 if (md->md_slice != sc->sc_slice) { 2674 G_MIRROR_DEBUG(1, 2675 "Invalid '%s' field on disk %s (device %s), skipping.", 2676 "md_slice", pp->name, sc->sc_name); 2677 return (EINVAL); 2678 } 2679 if (md->md_balance != sc->sc_balance) { 2680 G_MIRROR_DEBUG(1, 2681 "Invalid '%s' field on disk %s (device %s), skipping.", 2682 "md_balance", pp->name, sc->sc_name); 2683 return (EINVAL); 2684 } 2685 if (md->md_mediasize != sc->sc_mediasize) { 2686 G_MIRROR_DEBUG(1, 2687 "Invalid '%s' field on disk %s (device %s), skipping.", 2688 "md_mediasize", pp->name, sc->sc_name); 2689 return (EINVAL); 2690 } 2691 if (sc->sc_mediasize > pp->mediasize) { 2692 G_MIRROR_DEBUG(1, 2693 "Invalid size of disk %s (device %s), skipping.", pp->name, 2694 sc->sc_name); 2695 return (EINVAL); 2696 } 2697 if (md->md_sectorsize != sc->sc_sectorsize) { 2698 G_MIRROR_DEBUG(1, 2699 "Invalid '%s' field on disk %s (device %s), skipping.", 2700 "md_sectorsize", pp->name, sc->sc_name); 2701 return (EINVAL); 2702 } 2703 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2704 G_MIRROR_DEBUG(1, 2705 "Invalid sector size of disk %s (device %s), skipping.", 2706 pp->name, sc->sc_name); 2707 return (EINVAL); 2708 } 2709 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2710 G_MIRROR_DEBUG(1, 2711 "Invalid device flags on disk %s (device %s), skipping.", 2712 pp->name, sc->sc_name); 2713 return (EINVAL); 2714 } 2715 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2716 G_MIRROR_DEBUG(1, 2717 "Invalid disk flags on disk %s (device %s), skipping.", 2718 pp->name, sc->sc_name); 2719 return (EINVAL); 2720 } 2721 return (0); 2722} 2723 2724int 2725g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2726 struct g_mirror_metadata *md) 2727{ 2728 struct g_mirror_disk *disk; 2729 int error; 2730 2731 g_topology_assert_not(); 2732 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2733 2734 error = g_mirror_check_metadata(sc, pp, md); 2735 if (error != 0) 2736 return (error); 2737 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2738 md->md_genid < sc->sc_genid) { 2739 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2740 pp->name, sc->sc_name); 2741 return (EINVAL); 2742 } 2743 disk = g_mirror_init_disk(sc, pp, md, &error); 2744 if (disk == NULL) 2745 return (error); 2746 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2747 G_MIRROR_EVENT_WAIT); 2748 if (error != 0) 2749 return (error); 2750 if (md->md_version < G_MIRROR_VERSION) { 2751 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2752 pp->name, md->md_version, G_MIRROR_VERSION); 2753 g_mirror_update_metadata(disk); 2754 } 2755 return (0); 2756} 2757 2758static void 2759g_mirror_destroy_delayed(void *arg, int flag) 2760{ 2761 struct g_mirror_softc *sc; 2762 int error; 2763 2764 if (flag == EV_CANCEL) { 2765 G_MIRROR_DEBUG(1, "Destroying canceled."); 2766 return; 2767 } 2768 sc = arg; 2769 g_topology_unlock(); 2770 sx_xlock(&sc->sc_lock); 2771 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) == 0, 2772 ("DESTROY flag set on %s.", sc->sc_name)); 2773 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0, 2774 ("DESTROYING flag not set on %s.", sc->sc_name)); 2775 G_MIRROR_DEBUG(1, "Destroying %s (delayed).", sc->sc_name); 2776 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_SOFT); 2777 if (error != 0) { 2778 G_MIRROR_DEBUG(0, "Cannot destroy %s.", sc->sc_name); 2779 sx_xunlock(&sc->sc_lock); 2780 } 2781 g_topology_lock(); 2782} 2783 2784static int 2785g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2786{ 2787 struct g_mirror_softc *sc; 2788 int dcr, dcw, dce, error = 0; 2789 2790 g_topology_assert(); 2791 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2792 acw, ace); 2793 2794 sc = pp->geom->softc; 2795 if (sc == NULL && acr <= 0 && acw <= 0 && ace <= 0) 2796 return (0); 2797 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 2798 2799 dcr = pp->acr + acr; 2800 dcw = pp->acw + acw; 2801 dce = pp->ace + ace; 2802 2803 g_topology_unlock(); 2804 sx_xlock(&sc->sc_lock); 2805 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0 || 2806 LIST_EMPTY(&sc->sc_disks)) { 2807 if (acr > 0 || acw > 0 || ace > 0) 2808 error = ENXIO; 2809 goto end; 2810 } 2811 if (dcw == 0 && !sc->sc_idle) 2812 g_mirror_idle(sc, dcw); 2813 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0) { 2814 if (acr > 0 || acw > 0 || ace > 0) { 2815 error = ENXIO; 2816 goto end; 2817 } 2818 if (dcr == 0 && dcw == 0 && dce == 0) { 2819 g_post_event(g_mirror_destroy_delayed, sc, M_WAITOK, 2820 sc, NULL); 2821 } 2822 } 2823end: 2824 sx_xunlock(&sc->sc_lock); 2825 g_topology_lock(); 2826 return (error); 2827} 2828 2829static struct g_geom * 2830g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2831{ 2832 struct g_mirror_softc *sc; 2833 struct g_geom *gp; 2834 int error, timeout; 2835 2836 g_topology_assert(); 2837 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2838 md->md_mid); 2839 2840 /* One disk is minimum. */ 2841 if (md->md_all < 1) 2842 return (NULL); 2843 /* 2844 * Action geom. 2845 */ 2846 gp = g_new_geomf(mp, "%s", md->md_name); 2847 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2848 gp->start = g_mirror_start; 2849 gp->orphan = g_mirror_orphan; 2850 gp->access = g_mirror_access; 2851 gp->dumpconf = g_mirror_dumpconf; 2852 2853 sc->sc_id = md->md_mid; 2854 sc->sc_slice = md->md_slice; 2855 sc->sc_balance = md->md_balance; 2856 sc->sc_mediasize = md->md_mediasize; 2857 sc->sc_sectorsize = md->md_sectorsize; 2858 sc->sc_ndisks = md->md_all; 2859 sc->sc_flags = md->md_mflags; 2860 sc->sc_bump_id = 0; 2861 sc->sc_idle = 1; 2862 sc->sc_last_write = time_uptime; 2863 sc->sc_writes = 0; 2864 sx_init(&sc->sc_lock, "gmirror:lock"); 2865 bioq_init(&sc->sc_queue); 2866 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2867 bioq_init(&sc->sc_regular_delayed); 2868 bioq_init(&sc->sc_inflight); 2869 bioq_init(&sc->sc_sync_delayed); 2870 LIST_INIT(&sc->sc_disks); 2871 TAILQ_INIT(&sc->sc_events); 2872 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2873 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2874 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2875 gp->softc = sc; 2876 sc->sc_geom = gp; 2877 sc->sc_provider = NULL; 2878 /* 2879 * Synchronization geom. 2880 */ 2881 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2882 gp->softc = sc; 2883 gp->orphan = g_mirror_orphan; 2884 sc->sc_sync.ds_geom = gp; 2885 sc->sc_sync.ds_ndisks = 0; 2886 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2887 "g_mirror %s", md->md_name); 2888 if (error != 0) { 2889 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2890 sc->sc_name); 2891 g_destroy_geom(sc->sc_sync.ds_geom); 2892 mtx_destroy(&sc->sc_events_mtx); 2893 mtx_destroy(&sc->sc_queue_mtx); 2894 sx_destroy(&sc->sc_lock); 2895 g_destroy_geom(sc->sc_geom); 2896 free(sc, M_MIRROR); 2897 return (NULL); 2898 } 2899 2900 G_MIRROR_DEBUG(1, "Device %s created (%u components, id=%u).", 2901 sc->sc_name, sc->sc_ndisks, sc->sc_id); 2902 2903 sc->sc_rootmount = root_mount_hold("GMIRROR"); 2904 G_MIRROR_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); 2905 /* 2906 * Run timeout. 2907 */ 2908 timeout = g_mirror_timeout * hz; 2909 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 2910 return (sc->sc_geom); 2911} 2912 2913int 2914g_mirror_destroy(struct g_mirror_softc *sc, int how) 2915{ 2916 struct g_mirror_disk *disk; 2917 struct g_provider *pp; 2918 2919 g_topology_assert_not(); 2920 if (sc == NULL) 2921 return (ENXIO); 2922 sx_assert(&sc->sc_lock, SX_XLOCKED); 2923 2924 pp = sc->sc_provider; 2925 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2926 switch (how) { 2927 case G_MIRROR_DESTROY_SOFT: 2928 G_MIRROR_DEBUG(1, 2929 "Device %s is still open (r%dw%de%d).", pp->name, 2930 pp->acr, pp->acw, pp->ace); 2931 return (EBUSY); 2932 case G_MIRROR_DESTROY_DELAYED: 2933 G_MIRROR_DEBUG(1, 2934 "Device %s will be destroyed on last close.", 2935 pp->name); 2936 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2937 if (disk->d_state == 2938 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2939 g_mirror_sync_stop(disk, 1); 2940 } 2941 } 2942 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROYING; 2943 return (EBUSY); 2944 case G_MIRROR_DESTROY_HARD: 2945 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 2946 "can't be definitely removed.", pp->name); 2947 } 2948 } 2949 2950 g_topology_lock(); 2951 if (sc->sc_geom->softc == NULL) { 2952 g_topology_unlock(); 2953 return (0); 2954 } 2955 sc->sc_geom->softc = NULL; 2956 sc->sc_sync.ds_geom->softc = NULL; 2957 g_topology_unlock(); 2958 2959 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2960 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 2961 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2962 sx_xunlock(&sc->sc_lock); 2963 mtx_lock(&sc->sc_queue_mtx); 2964 wakeup(sc); 2965 mtx_unlock(&sc->sc_queue_mtx); 2966 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2967 while (sc->sc_worker != NULL) 2968 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 2969 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2970 sx_xlock(&sc->sc_lock); 2971 g_mirror_destroy_device(sc); 2972 free(sc, M_MIRROR); 2973 return (0); 2974} 2975 2976static void 2977g_mirror_taste_orphan(struct g_consumer *cp) 2978{ 2979 2980 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2981 cp->provider->name)); 2982} 2983 2984static struct g_geom * 2985g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2986{ 2987 struct g_mirror_metadata md; 2988 struct g_mirror_softc *sc; 2989 struct g_consumer *cp; 2990 struct g_geom *gp; 2991 int error; 2992 2993 g_topology_assert(); 2994 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2995 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 2996 2997 gp = g_new_geomf(mp, "mirror:taste"); 2998 /* 2999 * This orphan function should be never called. 3000 */ 3001 gp->orphan = g_mirror_taste_orphan; 3002 cp = g_new_consumer(gp); 3003 g_attach(cp, pp); 3004 error = g_mirror_read_metadata(cp, &md); 3005 g_detach(cp); 3006 g_destroy_consumer(cp); 3007 g_destroy_geom(gp); 3008 if (error != 0) 3009 return (NULL); 3010 gp = NULL; 3011 3012 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 3013 return (NULL); 3014 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 3015 return (NULL); 3016 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 3017 G_MIRROR_DEBUG(0, 3018 "Device %s: provider %s marked as inactive, skipping.", 3019 md.md_name, pp->name); 3020 return (NULL); 3021 } 3022 if (g_mirror_debug >= 2) 3023 mirror_metadata_dump(&md); 3024 3025 /* 3026 * Let's check if device already exists. 3027 */ 3028 sc = NULL; 3029 LIST_FOREACH(gp, &mp->geom, geom) { 3030 sc = gp->softc; 3031 if (sc == NULL) 3032 continue; 3033 if (sc->sc_sync.ds_geom == gp) 3034 continue; 3035 if (strcmp(md.md_name, sc->sc_name) != 0) 3036 continue; 3037 if (md.md_mid != sc->sc_id) { 3038 G_MIRROR_DEBUG(0, "Device %s already configured.", 3039 sc->sc_name); 3040 return (NULL); 3041 } 3042 break; 3043 } 3044 if (gp == NULL) { 3045 gp = g_mirror_create(mp, &md); 3046 if (gp == NULL) { 3047 G_MIRROR_DEBUG(0, "Cannot create device %s.", 3048 md.md_name); 3049 return (NULL); 3050 } 3051 sc = gp->softc; 3052 } 3053 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 3054 g_topology_unlock(); 3055 sx_xlock(&sc->sc_lock); 3056 error = g_mirror_add_disk(sc, pp, &md); 3057 if (error != 0) { 3058 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 3059 pp->name, gp->name, error); 3060 if (LIST_EMPTY(&sc->sc_disks)) { 3061 g_cancel_event(sc); 3062 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3063 g_topology_lock(); 3064 return (NULL); 3065 } 3066 gp = NULL; 3067 } 3068 sx_xunlock(&sc->sc_lock); 3069 g_topology_lock(); 3070 return (gp); 3071} 3072 3073static int 3074g_mirror_destroy_geom(struct gctl_req *req __unused, 3075 struct g_class *mp __unused, struct g_geom *gp) 3076{ 3077 struct g_mirror_softc *sc; 3078 int error; 3079 3080 g_topology_unlock(); 3081 sc = gp->softc; 3082 sx_xlock(&sc->sc_lock); 3083 g_cancel_event(sc); 3084 error = g_mirror_destroy(gp->softc, G_MIRROR_DESTROY_SOFT); 3085 if (error != 0) 3086 sx_xunlock(&sc->sc_lock); 3087 g_topology_lock(); 3088 return (error); 3089} 3090 3091static void 3092g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 3093 struct g_consumer *cp, struct g_provider *pp) 3094{ 3095 struct g_mirror_softc *sc; 3096 3097 g_topology_assert(); 3098 3099 sc = gp->softc; 3100 if (sc == NULL) 3101 return; 3102 /* Skip synchronization geom. */ 3103 if (gp == sc->sc_sync.ds_geom) 3104 return; 3105 if (pp != NULL) { 3106 /* Nothing here. */ 3107 } else if (cp != NULL) { 3108 struct g_mirror_disk *disk; 3109 3110 disk = cp->private; 3111 if (disk == NULL) 3112 return; 3113 g_topology_unlock(); 3114 sx_xlock(&sc->sc_lock); 3115 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 3116 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3117 sbuf_printf(sb, "%s<Synchronized>", indent); 3118 if (disk->d_sync.ds_offset == 0) 3119 sbuf_printf(sb, "0%%"); 3120 else { 3121 sbuf_printf(sb, "%u%%", 3122 (u_int)((disk->d_sync.ds_offset * 100) / 3123 sc->sc_provider->mediasize)); 3124 } 3125 sbuf_printf(sb, "</Synchronized>\n"); 3126 } 3127 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 3128 disk->d_sync.ds_syncid); 3129 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 3130 disk->d_genid); 3131 sbuf_printf(sb, "%s<Flags>", indent); 3132 if (disk->d_flags == 0) 3133 sbuf_printf(sb, "NONE"); 3134 else { 3135 int first = 1; 3136 3137#define ADD_FLAG(flag, name) do { \ 3138 if ((disk->d_flags & (flag)) != 0) { \ 3139 if (!first) \ 3140 sbuf_printf(sb, ", "); \ 3141 else \ 3142 first = 0; \ 3143 sbuf_printf(sb, name); \ 3144 } \ 3145} while (0) 3146 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 3147 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 3148 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 3149 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 3150 "SYNCHRONIZING"); 3151 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 3152 ADD_FLAG(G_MIRROR_DISK_FLAG_BROKEN, "BROKEN"); 3153#undef ADD_FLAG 3154 } 3155 sbuf_printf(sb, "</Flags>\n"); 3156 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 3157 disk->d_priority); 3158 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3159 g_mirror_disk_state2str(disk->d_state)); 3160 sx_xunlock(&sc->sc_lock); 3161 g_topology_lock(); 3162 } else { 3163 g_topology_unlock(); 3164 sx_xlock(&sc->sc_lock); 3165 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 3166 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 3167 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 3168 sbuf_printf(sb, "%s<Flags>", indent); 3169 if (sc->sc_flags == 0) 3170 sbuf_printf(sb, "NONE"); 3171 else { 3172 int first = 1; 3173 3174#define ADD_FLAG(flag, name) do { \ 3175 if ((sc->sc_flags & (flag)) != 0) { \ 3176 if (!first) \ 3177 sbuf_printf(sb, ", "); \ 3178 else \ 3179 first = 0; \ 3180 sbuf_printf(sb, name); \ 3181 } \ 3182} while (0) 3183 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 3184#undef ADD_FLAG 3185 } 3186 sbuf_printf(sb, "</Flags>\n"); 3187 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 3188 (u_int)sc->sc_slice); 3189 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 3190 balance_name(sc->sc_balance)); 3191 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 3192 sc->sc_ndisks); 3193 sbuf_printf(sb, "%s<State>", indent); 3194 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 3195 sbuf_printf(sb, "%s", "STARTING"); 3196 else if (sc->sc_ndisks == 3197 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 3198 sbuf_printf(sb, "%s", "COMPLETE"); 3199 else 3200 sbuf_printf(sb, "%s", "DEGRADED"); 3201 sbuf_printf(sb, "</State>\n"); 3202 sx_xunlock(&sc->sc_lock); 3203 g_topology_lock(); 3204 } 3205} 3206 3207static void 3208g_mirror_shutdown_pre_sync(void *arg, int howto) 3209{ 3210 struct g_class *mp; 3211 struct g_geom *gp, *gp2; 3212 struct g_mirror_softc *sc; 3213 int error; 3214 3215 mp = arg; 3216 DROP_GIANT(); 3217 g_topology_lock(); 3218 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3219 if ((sc = gp->softc) == NULL) 3220 continue; 3221 /* Skip synchronization geom. */ 3222 if (gp == sc->sc_sync.ds_geom) 3223 continue; 3224 g_topology_unlock(); 3225 sx_xlock(&sc->sc_lock); 3226 g_cancel_event(sc); 3227 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_DELAYED); 3228 if (error != 0) 3229 sx_xunlock(&sc->sc_lock); 3230 g_topology_lock(); 3231 } 3232 g_topology_unlock(); 3233 PICKUP_GIANT(); 3234} 3235 3236static void 3237g_mirror_init(struct g_class *mp) 3238{ 3239 3240 g_mirror_pre_sync = EVENTHANDLER_REGISTER(shutdown_pre_sync, 3241 g_mirror_shutdown_pre_sync, mp, SHUTDOWN_PRI_FIRST); 3242 if (g_mirror_pre_sync == NULL) 3243 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 3244} 3245 3246static void 3247g_mirror_fini(struct g_class *mp) 3248{ 3249 3250 if (g_mirror_pre_sync != NULL) 3251 EVENTHANDLER_DEREGISTER(shutdown_pre_sync, g_mirror_pre_sync); 3252} 3253 3254DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 3255